language:
- ru
license: cc-by-nc-sa-4.0
TL;DR: Text Normalization for Social Media Corpus
Dataset Description
This dataset contains examples of Russian-language texts from social networks with distorted spelling (typos, abbreviations, etc.) and their normalized versions in json format. A detailed spelling correction protocol is given in the TBA article. The dataset size is 1930 sentence pairs. In each pair, the sentences are tokenized by words, and the lengths of both sentences in the pair are equal. If a token had to be divided during the normalization process, the division point is marked with the "|" symbol. If two tokens were combined, the placeholder is set in place of the second token (or , if the final spelling contains a hyphen at the place of the combination). Punctuation was excluded during pre-processing. Dataset excerpt:
...
[
[
"Где",
"то",
"сейчас",
"один",
"сисадмин",
"Ищу",
"работу",
"опыт",
"работы",
"Google",
"YouTube"
],
[
"Где-то",
"<hyphen>",
"сейчас",
"один",
"системный|администратор",
"Ищу",
"работу",
"опыт",
"работы",
"<foreign>",
"<foreign>"
]
],
...
Source Data
Social networks subcorpus from the Russian National Corpus.
License
CC-BY-NC-SA-4.0
Citation
TBA